mirror of https://github.com/YosysHQ/yosys.git
1582 lines
39 KiB
C++
1582 lines
39 KiB
C++
/*
|
|
* ezSAT -- A simple and easy to use CNF generator for SAT solvers
|
|
*
|
|
* Copyright (C) 2013 Claire Xenia Wolf <claire@yosyshq.com>
|
|
*
|
|
* Permission to use, copy, modify, and/or distribute this software for any
|
|
* purpose with or without fee is hereby granted, provided that the above
|
|
* copyright notice and this permission notice appear in all copies.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
|
|
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
|
|
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
|
|
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
|
|
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
|
|
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
|
|
* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
|
|
*
|
|
*/
|
|
|
|
#include "ezsat.h"
|
|
|
|
#include <cmath>
|
|
#include <algorithm>
|
|
#include <cassert>
|
|
#include <string>
|
|
|
|
#include <stdlib.h>
|
|
|
|
const int ezSAT::CONST_TRUE = 1;
|
|
const int ezSAT::CONST_FALSE = 2;
|
|
|
|
static std::string my_int_to_string(int i)
|
|
{
|
|
#ifdef __MINGW32__
|
|
char buffer[64];
|
|
snprintf(buffer, 64, "%d", i);
|
|
return buffer;
|
|
#else
|
|
return std::to_string(i);
|
|
#endif
|
|
}
|
|
|
|
ezSAT::ezSAT()
|
|
{
|
|
statehash = 5381;
|
|
|
|
flag_keep_cnf = false;
|
|
flag_non_incremental = false;
|
|
|
|
non_incremental_solve_used_up = false;
|
|
|
|
cnfConsumed = false;
|
|
cnfVariableCount = 0;
|
|
cnfClausesCount = 0;
|
|
|
|
solverTimeout = 0;
|
|
solverTimoutStatus = false;
|
|
|
|
literal("CONST_TRUE");
|
|
literal("CONST_FALSE");
|
|
|
|
assert(literal("CONST_TRUE") == CONST_TRUE);
|
|
assert(literal("CONST_FALSE") == CONST_FALSE);
|
|
}
|
|
|
|
ezSAT::~ezSAT()
|
|
{
|
|
}
|
|
|
|
void ezSAT::addhash(unsigned int h)
|
|
{
|
|
statehash = ((statehash << 5) + statehash) ^ h;
|
|
}
|
|
|
|
int ezSAT::value(bool val)
|
|
{
|
|
return val ? CONST_TRUE : CONST_FALSE;
|
|
}
|
|
|
|
int ezSAT::literal()
|
|
{
|
|
literals.push_back(std::string());
|
|
return literals.size();
|
|
}
|
|
|
|
int ezSAT::literal(const std::string &name)
|
|
{
|
|
if (literalsCache.count(name) == 0) {
|
|
literals.push_back(name);
|
|
literalsCache[name] = literals.size();
|
|
}
|
|
return literalsCache.at(name);
|
|
}
|
|
|
|
int ezSAT::frozen_literal()
|
|
{
|
|
int id = literal();
|
|
freeze(id);
|
|
return id;
|
|
}
|
|
|
|
int ezSAT::frozen_literal(const std::string &name)
|
|
{
|
|
int id = literal(name);
|
|
freeze(id);
|
|
return id;
|
|
}
|
|
|
|
int ezSAT::expression(OpId op, int a, int b, int c, int d, int e, int f)
|
|
{
|
|
std::vector<int> args(6);
|
|
args[0] = a, args[1] = b, args[2] = c;
|
|
args[3] = d, args[4] = e, args[5] = f;
|
|
return expression(op, args);
|
|
}
|
|
|
|
int ezSAT::expression(OpId op, const std::vector<int> &args)
|
|
{
|
|
std::vector<int> myArgs;
|
|
myArgs.reserve(args.size());
|
|
bool xorRemovedOddTrues = false;
|
|
|
|
addhash(__LINE__);
|
|
addhash(op);
|
|
|
|
for (auto arg : args)
|
|
{
|
|
addhash(__LINE__);
|
|
addhash(arg);
|
|
|
|
if (arg == 0)
|
|
continue;
|
|
if (op == OpAnd && arg == CONST_TRUE)
|
|
continue;
|
|
if ((op == OpOr || op == OpXor) && arg == CONST_FALSE)
|
|
continue;
|
|
if (op == OpXor && arg == CONST_TRUE) {
|
|
xorRemovedOddTrues = !xorRemovedOddTrues;
|
|
continue;
|
|
}
|
|
myArgs.push_back(arg);
|
|
}
|
|
|
|
if (myArgs.size() > 0 && (op == OpAnd || op == OpOr || op == OpXor || op == OpIFF)) {
|
|
std::sort(myArgs.begin(), myArgs.end());
|
|
int j = 0;
|
|
for (int i = 1; i < int(myArgs.size()); i++)
|
|
if (j < 0 || myArgs[j] != myArgs[i])
|
|
myArgs[++j] = myArgs[i];
|
|
else if (op == OpXor)
|
|
j--;
|
|
myArgs.resize(j+1);
|
|
}
|
|
|
|
switch (op)
|
|
{
|
|
case OpNot:
|
|
assert(myArgs.size() == 1);
|
|
if (myArgs[0] == CONST_TRUE)
|
|
return CONST_FALSE;
|
|
if (myArgs[0] == CONST_FALSE)
|
|
return CONST_TRUE;
|
|
break;
|
|
|
|
case OpAnd:
|
|
if (myArgs.size() == 0)
|
|
return CONST_TRUE;
|
|
if (myArgs.size() == 1)
|
|
return myArgs[0];
|
|
break;
|
|
|
|
case OpOr:
|
|
if (myArgs.size() == 0)
|
|
return CONST_FALSE;
|
|
if (myArgs.size() == 1)
|
|
return myArgs[0];
|
|
break;
|
|
|
|
case OpXor:
|
|
if (myArgs.size() == 0)
|
|
return xorRemovedOddTrues ? CONST_TRUE : CONST_FALSE;
|
|
if (myArgs.size() == 1)
|
|
return xorRemovedOddTrues ? NOT(myArgs[0]) : myArgs[0];
|
|
break;
|
|
|
|
case OpIFF:
|
|
assert(myArgs.size() >= 1);
|
|
if (myArgs.size() == 1)
|
|
return CONST_TRUE;
|
|
// FIXME: Add proper const folding
|
|
break;
|
|
|
|
case OpITE:
|
|
assert(myArgs.size() == 3);
|
|
if (myArgs[0] == CONST_TRUE)
|
|
return myArgs[1];
|
|
if (myArgs[0] == CONST_FALSE)
|
|
return myArgs[2];
|
|
break;
|
|
|
|
default:
|
|
abort();
|
|
}
|
|
|
|
std::pair<OpId, std::vector<int>> myExpr(op, myArgs);
|
|
int id = 0;
|
|
|
|
if (expressionsCache.count(myExpr) > 0) {
|
|
id = expressionsCache.at(myExpr);
|
|
} else {
|
|
id = -(int(expressions.size()) + 1);
|
|
expressionsCache[myExpr] = id;
|
|
expressions.push_back(myExpr);
|
|
}
|
|
|
|
if (xorRemovedOddTrues)
|
|
id = NOT(id);
|
|
|
|
addhash(__LINE__);
|
|
addhash(id);
|
|
|
|
return id;
|
|
}
|
|
|
|
void ezSAT::lookup_literal(int id, std::string &name) const
|
|
{
|
|
assert(0 < id && id <= int(literals.size()));
|
|
name = literals[id - 1];
|
|
}
|
|
|
|
const std::string &ezSAT::lookup_literal(int id) const
|
|
{
|
|
assert(0 < id && id <= int(literals.size()));
|
|
return literals[id - 1];
|
|
}
|
|
|
|
void ezSAT::lookup_expression(int id, OpId &op, std::vector<int> &args) const
|
|
{
|
|
assert(0 < -id && -id <= int(expressions.size()));
|
|
op = expressions[-id - 1].first;
|
|
args = expressions[-id - 1].second;
|
|
}
|
|
|
|
const std::vector<int> &ezSAT::lookup_expression(int id, OpId &op) const
|
|
{
|
|
assert(0 < -id && -id <= int(expressions.size()));
|
|
op = expressions[-id - 1].first;
|
|
return expressions[-id - 1].second;
|
|
}
|
|
|
|
int ezSAT::parse_string(const std::string &)
|
|
{
|
|
abort();
|
|
}
|
|
|
|
std::string ezSAT::to_string(int id) const
|
|
{
|
|
std::string text;
|
|
|
|
if (id > 0)
|
|
{
|
|
lookup_literal(id, text);
|
|
}
|
|
else
|
|
{
|
|
OpId op;
|
|
std::vector<int> args;
|
|
lookup_expression(id, op, args);
|
|
|
|
switch (op)
|
|
{
|
|
case OpNot:
|
|
text = "not(";
|
|
break;
|
|
|
|
case OpAnd:
|
|
text = "and(";
|
|
break;
|
|
|
|
case OpOr:
|
|
text = "or(";
|
|
break;
|
|
|
|
case OpXor:
|
|
text = "xor(";
|
|
break;
|
|
|
|
case OpIFF:
|
|
text = "iff(";
|
|
break;
|
|
|
|
case OpITE:
|
|
text = "ite(";
|
|
break;
|
|
|
|
default:
|
|
abort();
|
|
}
|
|
|
|
for (int i = 0; i < int(args.size()); i++) {
|
|
if (i > 0)
|
|
text += ", ";
|
|
text += to_string(args[i]);
|
|
}
|
|
|
|
text += ")";
|
|
}
|
|
|
|
return text;
|
|
}
|
|
|
|
int ezSAT::eval(int id, const std::vector<int> &values) const
|
|
{
|
|
if (id > 0) {
|
|
if (id <= int(values.size()) && (values[id-1] == CONST_TRUE || values[id-1] == CONST_FALSE || values[id-1] == 0))
|
|
return values[id-1];
|
|
return 0;
|
|
}
|
|
|
|
OpId op;
|
|
const std::vector<int> &args = lookup_expression(id, op);
|
|
int a, b;
|
|
|
|
switch (op)
|
|
{
|
|
case OpNot:
|
|
assert(args.size() == 1);
|
|
a = eval(args[0], values);
|
|
if (a == CONST_TRUE)
|
|
return CONST_FALSE;
|
|
if (a == CONST_FALSE)
|
|
return CONST_TRUE;
|
|
return 0;
|
|
case OpAnd:
|
|
a = CONST_TRUE;
|
|
for (auto arg : args) {
|
|
b = eval(arg, values);
|
|
if (b != CONST_TRUE && b != CONST_FALSE)
|
|
a = 0;
|
|
if (b == CONST_FALSE)
|
|
return CONST_FALSE;
|
|
}
|
|
return a;
|
|
case OpOr:
|
|
a = CONST_FALSE;
|
|
for (auto arg : args) {
|
|
b = eval(arg, values);
|
|
if (b != CONST_TRUE && b != CONST_FALSE)
|
|
a = 0;
|
|
if (b == CONST_TRUE)
|
|
return CONST_TRUE;
|
|
}
|
|
return a;
|
|
case OpXor:
|
|
a = CONST_FALSE;
|
|
for (auto arg : args) {
|
|
b = eval(arg, values);
|
|
if (b != CONST_TRUE && b != CONST_FALSE)
|
|
return 0;
|
|
if (b == CONST_TRUE)
|
|
a = a == CONST_TRUE ? CONST_FALSE : CONST_TRUE;
|
|
}
|
|
return a;
|
|
case OpIFF:
|
|
assert(args.size() > 0);
|
|
a = eval(args[0], values);
|
|
for (auto arg : args) {
|
|
b = eval(arg, values);
|
|
if (b != CONST_TRUE && b != CONST_FALSE)
|
|
return 0;
|
|
if (b != a)
|
|
return CONST_FALSE;
|
|
}
|
|
return CONST_TRUE;
|
|
case OpITE:
|
|
assert(args.size() == 3);
|
|
a = eval(args[0], values);
|
|
if (a == CONST_TRUE)
|
|
return eval(args[1], values);
|
|
if (a == CONST_FALSE)
|
|
return eval(args[2], values);
|
|
return 0;
|
|
default:
|
|
abort();
|
|
}
|
|
}
|
|
|
|
void ezSAT::clear()
|
|
{
|
|
cnfConsumed = false;
|
|
cnfVariableCount = 0;
|
|
cnfClausesCount = 0;
|
|
cnfLiteralVariables.clear();
|
|
cnfExpressionVariables.clear();
|
|
cnfClauses.clear();
|
|
}
|
|
|
|
void ezSAT::freeze(int)
|
|
{
|
|
}
|
|
|
|
bool ezSAT::eliminated(int)
|
|
{
|
|
return false;
|
|
}
|
|
|
|
void ezSAT::assume(int id)
|
|
{
|
|
addhash(__LINE__);
|
|
addhash(id);
|
|
|
|
if (id < 0)
|
|
{
|
|
assert(0 < -id && -id <= int(expressions.size()));
|
|
cnfExpressionVariables.resize(expressions.size());
|
|
|
|
if (cnfExpressionVariables[-id-1] == 0)
|
|
{
|
|
OpId op;
|
|
std::vector<int> args;
|
|
lookup_expression(id, op, args);
|
|
|
|
if (op == OpNot) {
|
|
int idx = bind(args[0]);
|
|
cnfClauses.push_back(std::vector<int>(1, -idx));
|
|
cnfClausesCount++;
|
|
return;
|
|
}
|
|
if (op == OpOr) {
|
|
std::vector<int> clause;
|
|
for (int arg : args)
|
|
clause.push_back(bind(arg));
|
|
cnfClauses.push_back(clause);
|
|
cnfClausesCount++;
|
|
return;
|
|
}
|
|
if (op == OpAnd) {
|
|
for (int arg : args) {
|
|
cnfClauses.push_back(std::vector<int>(1, bind(arg)));
|
|
cnfClausesCount++;
|
|
}
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
|
|
int idx = bind(id);
|
|
cnfClauses.push_back(std::vector<int>(1, idx));
|
|
cnfClausesCount++;
|
|
}
|
|
|
|
void ezSAT::add_clause(const std::vector<int> &args)
|
|
{
|
|
addhash(__LINE__);
|
|
for (auto arg : args)
|
|
addhash(arg);
|
|
|
|
cnfClauses.push_back(args);
|
|
cnfClausesCount++;
|
|
}
|
|
|
|
void ezSAT::add_clause(const std::vector<int> &args, bool argsPolarity, int a, int b, int c)
|
|
{
|
|
std::vector<int> clause;
|
|
for (auto arg : args)
|
|
clause.push_back(argsPolarity ? +arg : -arg);
|
|
if (a != 0)
|
|
clause.push_back(a);
|
|
if (b != 0)
|
|
clause.push_back(b);
|
|
if (c != 0)
|
|
clause.push_back(c);
|
|
add_clause(clause);
|
|
}
|
|
|
|
void ezSAT::add_clause(int a, int b, int c)
|
|
{
|
|
std::vector<int> clause;
|
|
if (a != 0)
|
|
clause.push_back(a);
|
|
if (b != 0)
|
|
clause.push_back(b);
|
|
if (c != 0)
|
|
clause.push_back(c);
|
|
add_clause(clause);
|
|
}
|
|
|
|
int ezSAT::bind_cnf_not(const std::vector<int> &args)
|
|
{
|
|
assert(args.size() == 1);
|
|
return -args[0];
|
|
}
|
|
|
|
int ezSAT::bind_cnf_and(const std::vector<int> &args)
|
|
{
|
|
assert(args.size() >= 2);
|
|
|
|
int idx = ++cnfVariableCount;
|
|
add_clause(args, false, idx);
|
|
|
|
for (auto arg : args)
|
|
add_clause(-idx, arg);
|
|
|
|
return idx;
|
|
}
|
|
|
|
int ezSAT::bind_cnf_or(const std::vector<int> &args)
|
|
{
|
|
assert(args.size() >= 2);
|
|
|
|
int idx = ++cnfVariableCount;
|
|
add_clause(args, true, -idx);
|
|
|
|
for (auto arg : args)
|
|
add_clause(idx, -arg);
|
|
|
|
return idx;
|
|
}
|
|
|
|
int ezSAT::bound(int id) const
|
|
{
|
|
if (id > 0 && id <= int(cnfLiteralVariables.size()))
|
|
return cnfLiteralVariables[id-1];
|
|
if (-id > 0 && -id <= int(cnfExpressionVariables.size()))
|
|
return cnfExpressionVariables[-id-1];
|
|
return 0;
|
|
}
|
|
|
|
std::string ezSAT::cnfLiteralInfo(int idx) const
|
|
{
|
|
for (int i = 0; i < int(cnfLiteralVariables.size()); i++) {
|
|
if (cnfLiteralVariables[i] == idx)
|
|
return to_string(i+1);
|
|
if (cnfLiteralVariables[i] == -idx)
|
|
return "NOT " + to_string(i+1);
|
|
}
|
|
for (int i = 0; i < int(cnfExpressionVariables.size()); i++) {
|
|
if (cnfExpressionVariables[i] == idx)
|
|
return to_string(-i-1);
|
|
if (cnfExpressionVariables[i] == -idx)
|
|
return "NOT " + to_string(-i-1);
|
|
}
|
|
return "<unnamed>";
|
|
}
|
|
|
|
int ezSAT::bind(int id, bool auto_freeze)
|
|
{
|
|
addhash(__LINE__);
|
|
addhash(id);
|
|
addhash(auto_freeze);
|
|
|
|
if (id >= 0) {
|
|
assert(0 < id && id <= int(literals.size()));
|
|
cnfLiteralVariables.resize(literals.size());
|
|
if (eliminated(cnfLiteralVariables[id-1])) {
|
|
fprintf(stderr, "ezSAT: Missing freeze on literal `%s'.\n", to_string(id).c_str());
|
|
abort();
|
|
}
|
|
if (cnfLiteralVariables[id-1] == 0) {
|
|
cnfLiteralVariables[id-1] = ++cnfVariableCount;
|
|
if (id == CONST_TRUE)
|
|
add_clause(+cnfLiteralVariables[id-1]);
|
|
if (id == CONST_FALSE)
|
|
add_clause(-cnfLiteralVariables[id-1]);
|
|
}
|
|
return cnfLiteralVariables[id-1];
|
|
}
|
|
|
|
assert(0 < -id && -id <= int(expressions.size()));
|
|
cnfExpressionVariables.resize(expressions.size());
|
|
|
|
if (eliminated(cnfExpressionVariables[-id-1]))
|
|
{
|
|
cnfExpressionVariables[-id-1] = 0;
|
|
|
|
// this will recursively call bind(id). within the recursion
|
|
// the cnf is pre-set to 0. an idx is allocated there, then it
|
|
// is frozen, then it returns here with the new idx already set.
|
|
if (auto_freeze)
|
|
freeze(id);
|
|
}
|
|
|
|
if (cnfExpressionVariables[-id-1] == 0)
|
|
{
|
|
OpId op;
|
|
std::vector<int> args;
|
|
lookup_expression(id, op, args);
|
|
int idx = 0;
|
|
|
|
if (op == OpXor) {
|
|
while (args.size() > 1) {
|
|
std::vector<int> newArgs;
|
|
for (int i = 0; i < int(args.size()); i += 2)
|
|
if (i+1 == int(args.size())) {
|
|
newArgs.push_back(args[i]);
|
|
} else {
|
|
int sub1 = AND(args[i], NOT(args[i+1]));
|
|
int sub2 = AND(NOT(args[i]), args[i+1]);
|
|
newArgs.push_back(OR(sub1, sub2));
|
|
}
|
|
args.swap(newArgs);
|
|
}
|
|
idx = bind(args.at(0), false);
|
|
goto assign_idx;
|
|
}
|
|
|
|
if (op == OpIFF) {
|
|
std::vector<int> invArgs;
|
|
for (auto arg : args)
|
|
invArgs.push_back(NOT(arg));
|
|
int sub1 = expression(OpAnd, args);
|
|
int sub2 = expression(OpAnd, invArgs);
|
|
idx = bind(OR(sub1, sub2), false);
|
|
goto assign_idx;
|
|
}
|
|
|
|
if (op == OpITE) {
|
|
int sub1 = AND(args[0], args[1]);
|
|
int sub2 = AND(NOT(args[0]), args[2]);
|
|
idx = bind(OR(sub1, sub2), false);
|
|
goto assign_idx;
|
|
}
|
|
|
|
for (int i = 0; i < int(args.size()); i++)
|
|
args[i] = bind(args[i], false);
|
|
|
|
switch (op)
|
|
{
|
|
case OpNot: idx = bind_cnf_not(args); break;
|
|
case OpAnd: idx = bind_cnf_and(args); break;
|
|
case OpOr: idx = bind_cnf_or(args); break;
|
|
default: abort();
|
|
}
|
|
|
|
assign_idx:
|
|
assert(idx != 0);
|
|
cnfExpressionVariables[-id-1] = idx;
|
|
}
|
|
|
|
return cnfExpressionVariables[-id-1];
|
|
}
|
|
|
|
void ezSAT::consumeCnf()
|
|
{
|
|
if (mode_keep_cnf())
|
|
cnfClausesBackup.insert(cnfClausesBackup.end(), cnfClauses.begin(), cnfClauses.end());
|
|
else
|
|
cnfConsumed = true;
|
|
cnfClauses.clear();
|
|
}
|
|
|
|
void ezSAT::consumeCnf(std::vector<std::vector<int>> &cnf)
|
|
{
|
|
if (mode_keep_cnf())
|
|
cnfClausesBackup.insert(cnfClausesBackup.end(), cnfClauses.begin(), cnfClauses.end());
|
|
else
|
|
cnfConsumed = true;
|
|
cnf.swap(cnfClauses);
|
|
cnfClauses.clear();
|
|
}
|
|
|
|
void ezSAT::getFullCnf(std::vector<std::vector<int>> &full_cnf) const
|
|
{
|
|
assert(full_cnf.empty());
|
|
full_cnf.insert(full_cnf.end(), cnfClausesBackup.begin(), cnfClausesBackup.end());
|
|
full_cnf.insert(full_cnf.end(), cnfClauses.begin(), cnfClauses.end());
|
|
}
|
|
|
|
void ezSAT::preSolverCallback()
|
|
{
|
|
assert(!non_incremental_solve_used_up);
|
|
if (mode_non_incremental())
|
|
non_incremental_solve_used_up = true;
|
|
}
|
|
|
|
bool ezSAT::solver(const std::vector<int>&, std::vector<bool>&, const std::vector<int>&)
|
|
{
|
|
preSolverCallback();
|
|
fprintf(stderr, "************************************************************************\n");
|
|
fprintf(stderr, "ERROR: You are trying to use the solve() method of the ezSAT base class!\n");
|
|
fprintf(stderr, "Use a dervied class like ezMiniSAT instead.\n");
|
|
fprintf(stderr, "************************************************************************\n");
|
|
abort();
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_const(const std::vector<bool> &bits)
|
|
{
|
|
std::vector<int> vec;
|
|
for (auto bit : bits)
|
|
vec.push_back(bit ? CONST_TRUE : CONST_FALSE);
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_const_signed(int64_t value, int numBits)
|
|
{
|
|
std::vector<int> vec;
|
|
for (int i = 0; i < numBits; i++)
|
|
vec.push_back(((value >> i) & 1) != 0 ? CONST_TRUE : CONST_FALSE);
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_const_unsigned(uint64_t value, int numBits)
|
|
{
|
|
std::vector<int> vec;
|
|
for (int i = 0; i < numBits; i++)
|
|
vec.push_back(((value >> i) & 1) != 0 ? CONST_TRUE : CONST_FALSE);
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_var(int numBits)
|
|
{
|
|
std::vector<int> vec;
|
|
for (int i = 0; i < numBits; i++)
|
|
vec.push_back(literal());
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_var(std::string name, int numBits)
|
|
{
|
|
std::vector<int> vec;
|
|
for (int i = 0; i < numBits; i++) {
|
|
vec.push_back(VAR(name + my_int_to_string(i)));
|
|
}
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_cast(const std::vector<int> &vec1, int toBits, bool signExtend)
|
|
{
|
|
std::vector<int> vec;
|
|
for (int i = 0; i < toBits; i++)
|
|
if (i >= int(vec1.size()))
|
|
vec.push_back(signExtend ? vec1.back() : CONST_FALSE);
|
|
else
|
|
vec.push_back(vec1[i]);
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_not(const std::vector<int> &vec1)
|
|
{
|
|
std::vector<int> vec;
|
|
for (auto bit : vec1)
|
|
vec.push_back(NOT(bit));
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_and(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
assert(vec1.size() == vec2.size());
|
|
std::vector<int> vec(vec1.size());
|
|
for (int i = 0; i < int(vec1.size()); i++)
|
|
vec[i] = AND(vec1[i], vec2[i]);
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_or(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
assert(vec1.size() == vec2.size());
|
|
std::vector<int> vec(vec1.size());
|
|
for (int i = 0; i < int(vec1.size()); i++)
|
|
vec[i] = OR(vec1[i], vec2[i]);
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_xor(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
assert(vec1.size() == vec2.size());
|
|
std::vector<int> vec(vec1.size());
|
|
for (int i = 0; i < int(vec1.size()); i++)
|
|
vec[i] = XOR(vec1[i], vec2[i]);
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_iff(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
assert(vec1.size() == vec2.size());
|
|
std::vector<int> vec(vec1.size());
|
|
for (int i = 0; i < int(vec1.size()); i++)
|
|
vec[i] = IFF(vec1[i], vec2[i]);
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_ite(const std::vector<int> &vec1, const std::vector<int> &vec2, const std::vector<int> &vec3)
|
|
{
|
|
assert(vec1.size() == vec2.size() && vec2.size() == vec3.size());
|
|
std::vector<int> vec(vec1.size());
|
|
for (int i = 0; i < int(vec1.size()); i++)
|
|
vec[i] = ITE(vec1[i], vec2[i], vec3[i]);
|
|
return vec;
|
|
}
|
|
|
|
|
|
std::vector<int> ezSAT::vec_ite(int sel, const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
assert(vec1.size() == vec2.size());
|
|
std::vector<int> vec(vec1.size());
|
|
for (int i = 0; i < int(vec1.size()); i++)
|
|
vec[i] = ITE(sel, vec1[i], vec2[i]);
|
|
return vec;
|
|
}
|
|
|
|
// 'y' is the MSB (carry) and x the LSB (sum) output
|
|
static void fulladder(ezSAT *that, int a, int b, int c, int &y, int &x)
|
|
{
|
|
int tmp = that->XOR(a, b);
|
|
int new_x = that->XOR(tmp, c);
|
|
int new_y = that->OR(that->AND(a, b), that->AND(c, tmp));
|
|
#if 0
|
|
printf("FULLADD> a=%s, b=%s, c=%s, carry=%s, sum=%s\n", that->to_string(a).c_str(), that->to_string(b).c_str(),
|
|
that->to_string(c).c_str(), that->to_string(new_y).c_str(), that->to_string(new_x).c_str());
|
|
#endif
|
|
x = new_x, y = new_y;
|
|
}
|
|
|
|
// 'y' is the MSB (carry) and x the LSB (sum) output
|
|
static void halfadder(ezSAT *that, int a, int b, int &y, int &x)
|
|
{
|
|
int new_x = that->XOR(a, b);
|
|
int new_y = that->AND(a, b);
|
|
#if 0
|
|
printf("HALFADD> a=%s, b=%s, carry=%s, sum=%s\n", that->to_string(a).c_str(), that->to_string(b).c_str(),
|
|
that->to_string(new_y).c_str(), that->to_string(new_x).c_str());
|
|
#endif
|
|
x = new_x, y = new_y;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_count(const std::vector<int> &vec, int numBits, bool clip)
|
|
{
|
|
std::vector<int> sum = vec_const_unsigned(0, numBits);
|
|
std::vector<int> carry_vector;
|
|
|
|
for (auto bit : vec) {
|
|
int carry = bit;
|
|
for (int i = 0; i < numBits; i++)
|
|
halfadder(this, carry, sum[i], carry, sum[i]);
|
|
carry_vector.push_back(carry);
|
|
}
|
|
|
|
if (clip) {
|
|
int overflow = vec_reduce_or(carry_vector);
|
|
sum = vec_ite(overflow, vec_const_unsigned(~0, numBits), sum);
|
|
}
|
|
|
|
#if 0
|
|
printf("COUNT> vec=[");
|
|
for (int i = int(vec.size())-1; i >= 0; i--)
|
|
printf("%s%s", to_string(vec[i]).c_str(), i ? ", " : "");
|
|
printf("], result=[");
|
|
for (int i = int(sum.size())-1; i >= 0; i--)
|
|
printf("%s%s", to_string(sum[i]).c_str(), i ? ", " : "");
|
|
printf("]\n");
|
|
#endif
|
|
|
|
return sum;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_add(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
assert(vec1.size() == vec2.size());
|
|
std::vector<int> vec(vec1.size());
|
|
int carry = CONST_FALSE;
|
|
for (int i = 0; i < int(vec1.size()); i++)
|
|
fulladder(this, vec1[i], vec2[i], carry, carry, vec[i]);
|
|
|
|
#if 0
|
|
printf("ADD> vec1=[");
|
|
for (int i = int(vec1.size())-1; i >= 0; i--)
|
|
printf("%s%s", to_string(vec1[i]).c_str(), i ? ", " : "");
|
|
printf("], vec2=[");
|
|
for (int i = int(vec2.size())-1; i >= 0; i--)
|
|
printf("%s%s", to_string(vec2[i]).c_str(), i ? ", " : "");
|
|
printf("], result=[");
|
|
for (int i = int(vec.size())-1; i >= 0; i--)
|
|
printf("%s%s", to_string(vec[i]).c_str(), i ? ", " : "");
|
|
printf("]\n");
|
|
#endif
|
|
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_sub(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
assert(vec1.size() == vec2.size());
|
|
std::vector<int> vec(vec1.size());
|
|
int carry = CONST_TRUE;
|
|
for (int i = 0; i < int(vec1.size()); i++)
|
|
fulladder(this, vec1[i], NOT(vec2[i]), carry, carry, vec[i]);
|
|
|
|
#if 0
|
|
printf("SUB> vec1=[");
|
|
for (int i = int(vec1.size())-1; i >= 0; i--)
|
|
printf("%s%s", to_string(vec1[i]).c_str(), i ? ", " : "");
|
|
printf("], vec2=[");
|
|
for (int i = int(vec2.size())-1; i >= 0; i--)
|
|
printf("%s%s", to_string(vec2[i]).c_str(), i ? ", " : "");
|
|
printf("], result=[");
|
|
for (int i = int(vec.size())-1; i >= 0; i--)
|
|
printf("%s%s", to_string(vec[i]).c_str(), i ? ", " : "");
|
|
printf("]\n");
|
|
#endif
|
|
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_neg(const std::vector<int> &vec)
|
|
{
|
|
std::vector<int> zero(vec.size(), CONST_FALSE);
|
|
return vec_sub(zero, vec);
|
|
}
|
|
|
|
void ezSAT::vec_cmp(const std::vector<int> &vec1, const std::vector<int> &vec2, int &carry, int &overflow, int &sign, int &zero)
|
|
{
|
|
assert(vec1.size() == vec2.size());
|
|
carry = CONST_TRUE;
|
|
zero = CONST_FALSE;
|
|
for (int i = 0; i < int(vec1.size()); i++) {
|
|
overflow = carry;
|
|
fulladder(this, vec1[i], NOT(vec2[i]), carry, carry, sign);
|
|
zero = OR(zero, sign);
|
|
}
|
|
overflow = XOR(overflow, carry);
|
|
carry = NOT(carry);
|
|
zero = NOT(zero);
|
|
|
|
#if 0
|
|
printf("CMP> vec1=[");
|
|
for (int i = int(vec1.size())-1; i >= 0; i--)
|
|
printf("%s%s", to_string(vec1[i]).c_str(), i ? ", " : "");
|
|
printf("], vec2=[");
|
|
for (int i = int(vec2.size())-1; i >= 0; i--)
|
|
printf("%s%s", to_string(vec2[i]).c_str(), i ? ", " : "");
|
|
printf("], carry=%s, overflow=%s, sign=%s, zero=%s\n", to_string(carry).c_str(), to_string(overflow).c_str(), to_string(sign).c_str(), to_string(zero).c_str());
|
|
#endif
|
|
}
|
|
|
|
int ezSAT::vec_lt_signed(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
int carry, overflow, sign, zero;
|
|
vec_cmp(vec1, vec2, carry, overflow, sign, zero);
|
|
return OR(AND(NOT(overflow), sign), AND(overflow, NOT(sign)));
|
|
}
|
|
|
|
int ezSAT::vec_le_signed(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
int carry, overflow, sign, zero;
|
|
vec_cmp(vec1, vec2, carry, overflow, sign, zero);
|
|
return OR(AND(NOT(overflow), sign), AND(overflow, NOT(sign)), zero);
|
|
}
|
|
|
|
int ezSAT::vec_ge_signed(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
int carry, overflow, sign, zero;
|
|
vec_cmp(vec1, vec2, carry, overflow, sign, zero);
|
|
return OR(AND(NOT(overflow), NOT(sign)), AND(overflow, sign));
|
|
}
|
|
|
|
int ezSAT::vec_gt_signed(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
int carry, overflow, sign, zero;
|
|
vec_cmp(vec1, vec2, carry, overflow, sign, zero);
|
|
return AND(OR(AND(NOT(overflow), NOT(sign)), AND(overflow, sign)), NOT(zero));
|
|
}
|
|
|
|
int ezSAT::vec_lt_unsigned(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
int carry, overflow, sign, zero;
|
|
vec_cmp(vec1, vec2, carry, overflow, sign, zero);
|
|
return carry;
|
|
}
|
|
|
|
int ezSAT::vec_le_unsigned(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
int carry, overflow, sign, zero;
|
|
vec_cmp(vec1, vec2, carry, overflow, sign, zero);
|
|
return OR(carry, zero);
|
|
}
|
|
|
|
int ezSAT::vec_ge_unsigned(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
int carry, overflow, sign, zero;
|
|
vec_cmp(vec1, vec2, carry, overflow, sign, zero);
|
|
return NOT(carry);
|
|
}
|
|
|
|
int ezSAT::vec_gt_unsigned(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
int carry, overflow, sign, zero;
|
|
vec_cmp(vec1, vec2, carry, overflow, sign, zero);
|
|
return AND(NOT(carry), NOT(zero));
|
|
}
|
|
|
|
int ezSAT::vec_eq(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
return vec_reduce_and(vec_iff(vec1, vec2));
|
|
}
|
|
|
|
int ezSAT::vec_ne(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
return NOT(vec_reduce_and(vec_iff(vec1, vec2)));
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_shl(const std::vector<int> &vec1, int shift, bool signExtend)
|
|
{
|
|
std::vector<int> vec;
|
|
for (int i = 0; i < int(vec1.size()); i++) {
|
|
int j = i-shift;
|
|
if (int(vec1.size()) <= j)
|
|
vec.push_back(signExtend ? vec1.back() : CONST_FALSE);
|
|
else if (0 <= j)
|
|
vec.push_back(vec1[j]);
|
|
else
|
|
vec.push_back(CONST_FALSE);
|
|
}
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_srl(const std::vector<int> &vec1, int shift)
|
|
{
|
|
std::vector<int> vec;
|
|
for (int i = 0; i < int(vec1.size()); i++) {
|
|
int j = i-shift;
|
|
while (j < 0)
|
|
j += vec1.size();
|
|
while (j >= int(vec1.size()))
|
|
j -= vec1.size();
|
|
vec.push_back(vec1[j]);
|
|
}
|
|
return vec;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_shift(const std::vector<int> &vec1, int shift, int extend_left, int extend_right)
|
|
{
|
|
std::vector<int> vec;
|
|
for (int i = 0; i < int(vec1.size()); i++) {
|
|
int j = i+shift;
|
|
if (j < 0)
|
|
vec.push_back(extend_right);
|
|
else if (j >= int(vec1.size()))
|
|
vec.push_back(extend_left);
|
|
else
|
|
vec.push_back(vec1[j]);
|
|
}
|
|
return vec;
|
|
}
|
|
|
|
static int my_clog2(int x)
|
|
{
|
|
int result = 0;
|
|
for (x--; x > 0; result++)
|
|
x >>= 1;
|
|
return result;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_shift_right(const std::vector<int> &vec1, const std::vector<int> &vec2, bool vec2_signed, int extend_left, int extend_right)
|
|
{
|
|
int vec2_bits = std::min(my_clog2(vec1.size()) + (vec2_signed ? 1 : 0), int(vec2.size()));
|
|
|
|
std::vector<int> overflow_bits(vec2.begin() + vec2_bits, vec2.end());
|
|
int overflow_left = CONST_FALSE, overflow_right = CONST_FALSE;
|
|
|
|
if (vec2_signed) {
|
|
int overflow = CONST_FALSE;
|
|
for (auto bit : overflow_bits)
|
|
overflow = OR(overflow, XOR(bit, vec2[vec2_bits-1]));
|
|
overflow_left = AND(overflow, NOT(vec2.back()));
|
|
overflow_right = AND(overflow, vec2.back());
|
|
} else
|
|
overflow_left = vec_reduce_or(overflow_bits);
|
|
|
|
std::vector<int> buffer = vec1;
|
|
|
|
if (vec2_signed)
|
|
while (buffer.size() < vec1.size() + (1 << vec2_bits))
|
|
buffer.push_back(extend_left);
|
|
|
|
std::vector<int> overflow_pattern_left(buffer.size(), extend_left);
|
|
std::vector<int> overflow_pattern_right(buffer.size(), extend_right);
|
|
|
|
buffer = vec_ite(overflow_left, overflow_pattern_left, buffer);
|
|
|
|
if (vec2_signed)
|
|
buffer = vec_ite(overflow_right, overflow_pattern_left, buffer);
|
|
|
|
for (int i = vec2_bits-1; i >= 0; i--) {
|
|
std::vector<int> shifted_buffer;
|
|
if (vec2_signed && i == vec2_bits-1)
|
|
shifted_buffer = vec_shift(buffer, -(1 << i), extend_left, extend_right);
|
|
else
|
|
shifted_buffer = vec_shift(buffer, 1 << i, extend_left, extend_right);
|
|
buffer = vec_ite(vec2[i], shifted_buffer, buffer);
|
|
}
|
|
|
|
buffer.resize(vec1.size());
|
|
return buffer;
|
|
}
|
|
|
|
std::vector<int> ezSAT::vec_shift_left(const std::vector<int> &vec1, const std::vector<int> &vec2, bool vec2_signed, int extend_left, int extend_right)
|
|
{
|
|
// vec2_signed is not implemented in vec_shift_left() yet
|
|
if (vec2_signed) assert(vec2_signed == false);
|
|
|
|
int vec2_bits = std::min(my_clog2(vec1.size()), int(vec2.size()));
|
|
|
|
std::vector<int> overflow_bits(vec2.begin() + vec2_bits, vec2.end());
|
|
int overflow = vec_reduce_or(overflow_bits);
|
|
|
|
std::vector<int> buffer = vec1;
|
|
std::vector<int> overflow_pattern_right(buffer.size(), extend_right);
|
|
buffer = vec_ite(overflow, overflow_pattern_right, buffer);
|
|
|
|
for (int i = 0; i < vec2_bits; i++) {
|
|
std::vector<int> shifted_buffer;
|
|
shifted_buffer = vec_shift(buffer, -(1 << i), extend_left, extend_right);
|
|
buffer = vec_ite(vec2[i], shifted_buffer, buffer);
|
|
}
|
|
|
|
buffer.resize(vec1.size());
|
|
return buffer;
|
|
}
|
|
|
|
void ezSAT::vec_append(std::vector<int> &vec, const std::vector<int> &vec1) const
|
|
{
|
|
for (auto bit : vec1)
|
|
vec.push_back(bit);
|
|
}
|
|
|
|
void ezSAT::vec_append_signed(std::vector<int> &vec, const std::vector<int> &vec1, int64_t value)
|
|
{
|
|
assert(int(vec1.size()) <= 64);
|
|
for (int i = 0; i < int(vec1.size()); i++) {
|
|
if (((value >> i) & 1) != 0)
|
|
vec.push_back(vec1[i]);
|
|
else
|
|
vec.push_back(NOT(vec1[i]));
|
|
}
|
|
}
|
|
|
|
void ezSAT::vec_append_unsigned(std::vector<int> &vec, const std::vector<int> &vec1, uint64_t value)
|
|
{
|
|
assert(int(vec1.size()) <= 64);
|
|
for (int i = 0; i < int(vec1.size()); i++) {
|
|
if (((value >> i) & 1) != 0)
|
|
vec.push_back(vec1[i]);
|
|
else
|
|
vec.push_back(NOT(vec1[i]));
|
|
}
|
|
}
|
|
|
|
int64_t ezSAT::vec_model_get_signed(const std::vector<int> &modelExpressions, const std::vector<bool> &modelValues, const std::vector<int> &vec1) const
|
|
{
|
|
int64_t value = 0;
|
|
std::map<int, bool> modelMap;
|
|
assert(modelExpressions.size() == modelValues.size());
|
|
for (int i = 0; i < int(modelExpressions.size()); i++)
|
|
modelMap[modelExpressions[i]] = modelValues[i];
|
|
for (int i = 0; i < 64; i++) {
|
|
int j = i < int(vec1.size()) ? i : vec1.size()-1;
|
|
if (modelMap.at(vec1[j]))
|
|
value |= int64_t(1) << i;
|
|
}
|
|
return value;
|
|
}
|
|
|
|
uint64_t ezSAT::vec_model_get_unsigned(const std::vector<int> &modelExpressions, const std::vector<bool> &modelValues, const std::vector<int> &vec1) const
|
|
{
|
|
uint64_t value = 0;
|
|
std::map<int, bool> modelMap;
|
|
assert(modelExpressions.size() == modelValues.size());
|
|
for (int i = 0; i < int(modelExpressions.size()); i++)
|
|
modelMap[modelExpressions[i]] = modelValues[i];
|
|
for (int i = 0; i < int(vec1.size()); i++)
|
|
if (modelMap.at(vec1[i]))
|
|
value |= uint64_t(1) << i;
|
|
return value;
|
|
}
|
|
|
|
int ezSAT::vec_reduce_and(const std::vector<int> &vec1)
|
|
{
|
|
return expression(OpAnd, vec1);
|
|
}
|
|
|
|
int ezSAT::vec_reduce_or(const std::vector<int> &vec1)
|
|
{
|
|
return expression(OpOr, vec1);
|
|
}
|
|
|
|
void ezSAT::vec_set(const std::vector<int> &vec1, const std::vector<int> &vec2)
|
|
{
|
|
assert(vec1.size() == vec2.size());
|
|
for (int i = 0; i < int(vec1.size()); i++)
|
|
SET(vec1[i], vec2[i]);
|
|
}
|
|
|
|
void ezSAT::vec_set_signed(const std::vector<int> &vec1, int64_t value)
|
|
{
|
|
assert(int(vec1.size()) <= 64);
|
|
for (int i = 0; i < int(vec1.size()); i++) {
|
|
if (((value >> i) & 1) != 0)
|
|
assume(vec1[i]);
|
|
else
|
|
assume(NOT(vec1[i]));
|
|
}
|
|
}
|
|
|
|
void ezSAT::vec_set_unsigned(const std::vector<int> &vec1, uint64_t value)
|
|
{
|
|
assert(int(vec1.size()) <= 64);
|
|
for (int i = 0; i < int(vec1.size()); i++) {
|
|
if (((value >> i) & 1) != 0)
|
|
assume(vec1[i]);
|
|
else
|
|
assume(NOT(vec1[i]));
|
|
}
|
|
}
|
|
|
|
ezSATbit ezSAT::bit(_V a)
|
|
{
|
|
return ezSATbit(*this, a);
|
|
}
|
|
|
|
ezSATvec ezSAT::vec(const std::vector<int> &vec)
|
|
{
|
|
return ezSATvec(*this, vec);
|
|
}
|
|
|
|
void ezSAT::printDIMACS(FILE *f, bool verbose) const
|
|
{
|
|
if (cnfConsumed) {
|
|
fprintf(stderr, "Usage error: printDIMACS() must not be called after cnfConsumed()!");
|
|
abort();
|
|
}
|
|
|
|
int digits = ceil(log10f(cnfVariableCount)) + 2;
|
|
|
|
fprintf(f, "c generated by ezSAT\n");
|
|
|
|
if (verbose)
|
|
{
|
|
fprintf(f, "c\n");
|
|
fprintf(f, "c mapping of variables to literals:\n");
|
|
for (int i = 0; i < int(cnfLiteralVariables.size()); i++)
|
|
if (cnfLiteralVariables[i] != 0)
|
|
fprintf(f, "c %*d: %s\n", digits, cnfLiteralVariables[i], literals[i].c_str());
|
|
|
|
fprintf(f, "c\n");
|
|
fprintf(f, "c mapping of variables to expressions:\n");
|
|
for (int i = 0; i < int(cnfExpressionVariables.size()); i++)
|
|
if (cnfExpressionVariables[i] != 0)
|
|
fprintf(f, "c %*d: %d\n", digits, cnfExpressionVariables[i], -i-1);
|
|
|
|
if (mode_keep_cnf()) {
|
|
fprintf(f, "c\n");
|
|
fprintf(f, "c %d clauses from backup, %d from current buffer\n",
|
|
int(cnfClausesBackup.size()), int(cnfClauses.size()));
|
|
}
|
|
|
|
fprintf(f, "c\n");
|
|
}
|
|
|
|
std::vector<std::vector<int>> all_clauses;
|
|
getFullCnf(all_clauses);
|
|
assert(cnfClausesCount == int(all_clauses.size()));
|
|
|
|
fprintf(f, "p cnf %d %d\n", cnfVariableCount, cnfClausesCount);
|
|
int maxClauseLen = 0;
|
|
for (auto &clause : all_clauses)
|
|
maxClauseLen = std::max(int(clause.size()), maxClauseLen);
|
|
if (!verbose)
|
|
maxClauseLen = std::min(maxClauseLen, 3);
|
|
for (auto &clause : all_clauses) {
|
|
for (auto idx : clause)
|
|
fprintf(f, " %*d", digits, idx);
|
|
if (maxClauseLen >= int(clause.size()))
|
|
fprintf(f, " %*d\n", (digits + 1)*int(maxClauseLen - clause.size()) + digits, 0);
|
|
else
|
|
fprintf(f, " %*d\n", digits, 0);
|
|
}
|
|
}
|
|
|
|
static std::string expression2str(const std::pair<ezSAT::OpId, std::vector<int>> &data)
|
|
{
|
|
std::string text;
|
|
switch (data.first) {
|
|
#define X(op) case ezSAT::op: text += #op; break;
|
|
X(OpNot)
|
|
X(OpAnd)
|
|
X(OpOr)
|
|
X(OpXor)
|
|
X(OpIFF)
|
|
X(OpITE)
|
|
default:
|
|
abort();
|
|
#undef X
|
|
}
|
|
text += ":";
|
|
for (auto it : data.second)
|
|
text += " " + my_int_to_string(it);
|
|
return text;
|
|
}
|
|
|
|
void ezSAT::printInternalState(FILE *f) const
|
|
{
|
|
fprintf(f, "--8<-- snip --8<--\n");
|
|
|
|
fprintf(f, "literalsCache:\n");
|
|
for (auto &it : literalsCache)
|
|
fprintf(f, " `%s' -> %d\n", it.first.c_str(), it.second);
|
|
|
|
fprintf(f, "literals:\n");
|
|
for (int i = 0; i < int(literals.size()); i++)
|
|
fprintf(f, " %d: `%s'\n", i+1, literals[i].c_str());
|
|
|
|
fprintf(f, "expressionsCache:\n");
|
|
for (auto &it : expressionsCache)
|
|
fprintf(f, " `%s' -> %d\n", expression2str(it.first).c_str(), it.second);
|
|
|
|
fprintf(f, "expressions:\n");
|
|
for (int i = 0; i < int(expressions.size()); i++)
|
|
fprintf(f, " %d: `%s'\n", -i-1, expression2str(expressions[i]).c_str());
|
|
|
|
fprintf(f, "cnfVariables (count=%d):\n", cnfVariableCount);
|
|
for (int i = 0; i < int(cnfLiteralVariables.size()); i++)
|
|
if (cnfLiteralVariables[i] != 0)
|
|
fprintf(f, " literal %d -> %d (%s)\n", i+1, cnfLiteralVariables[i], to_string(i+1).c_str());
|
|
for (int i = 0; i < int(cnfExpressionVariables.size()); i++)
|
|
if (cnfExpressionVariables[i] != 0)
|
|
fprintf(f, " expression %d -> %d (%s)\n", -i-1, cnfExpressionVariables[i], to_string(-i-1).c_str());
|
|
|
|
fprintf(f, "cnfClauses:\n");
|
|
for (auto &i1 : cnfClauses) {
|
|
for (auto &i2 : i1)
|
|
fprintf(f, " %4d", i2);
|
|
fprintf(f, "\n");
|
|
}
|
|
if (cnfConsumed)
|
|
fprintf(f, " *** more clauses consumed via cnfConsume() ***\n");
|
|
|
|
fprintf(f, "--8<-- snap --8<--\n");
|
|
}
|
|
|
|
static int clog2(int x)
|
|
{
|
|
int y = (x & (x - 1));
|
|
y = (y | -y) >> 31;
|
|
|
|
x |= (x >> 1);
|
|
x |= (x >> 2);
|
|
x |= (x >> 4);
|
|
x |= (x >> 8);
|
|
x |= (x >> 16);
|
|
|
|
x >>= 1;
|
|
x -= ((x >> 1) & 0x55555555);
|
|
x = (((x >> 2) & 0x33333333) + (x & 0x33333333));
|
|
x = (((x >> 4) + x) & 0x0f0f0f0f);
|
|
x += (x >> 8);
|
|
x += (x >> 16);
|
|
x = x & 0x0000003f;
|
|
|
|
return x - y;
|
|
}
|
|
|
|
int ezSAT::onehot(const std::vector<int> &vec, bool max_only)
|
|
{
|
|
// Mixed one-hot/binary encoding as described by Claessen in Sec. 4.2 of
|
|
// "Successful SAT Encoding Techniques. Magnus Bjiirk. 25th July 2009".
|
|
// http://jsat.ewi.tudelft.nl/addendum/Bjork_encoding.pdf
|
|
|
|
std::vector<int> formula;
|
|
|
|
// add at-leat-one constraint
|
|
if (max_only == false)
|
|
formula.push_back(expression(OpOr, vec));
|
|
|
|
if (vec.size() < 8)
|
|
{
|
|
// fall-back to simple O(n^2) solution for small cases
|
|
for (size_t i = 0; i < vec.size(); i++)
|
|
for (size_t j = i+1; j < vec.size(); j++) {
|
|
std::vector<int> clause;
|
|
clause.push_back(NOT(vec[i]));
|
|
clause.push_back(NOT(vec[j]));
|
|
formula.push_back(expression(OpOr, clause));
|
|
}
|
|
}
|
|
else
|
|
{
|
|
// create binary vector
|
|
int num_bits = clog2(vec.size());
|
|
std::vector<int> bits;
|
|
for (int k = 0; k < num_bits; k++)
|
|
bits.push_back(literal());
|
|
|
|
// add at-most-one clauses using binary encoding
|
|
for (size_t i = 0; i < vec.size(); i++)
|
|
for (int k = 0; k < num_bits; k++) {
|
|
std::vector<int> clause;
|
|
clause.push_back(NOT(vec[i]));
|
|
clause.push_back((i & (1 << k)) != 0 ? bits[k] : NOT(bits[k]));
|
|
formula.push_back(expression(OpOr, clause));
|
|
}
|
|
}
|
|
|
|
return expression(OpAnd, formula);
|
|
}
|
|
|
|
#if 0
|
|
int ezSAT::manyhot(const std::vector<int> &vec, int min_hot, int max_hot)
|
|
{
|
|
// many-hot encoding using a simple sorting network
|
|
|
|
if (max_hot < 0)
|
|
max_hot = min_hot;
|
|
|
|
std::vector<int> formula;
|
|
int M = max_hot+1, N = vec.size();
|
|
std::map<std::pair<int,int>, int> x;
|
|
|
|
for (int i = -1; i < N; i++)
|
|
for (int j = -1; j < M; j++)
|
|
x[std::pair<int,int>(i,j)] = j < 0 ? CONST_TRUE : i < 0 ? CONST_FALSE : literal();
|
|
|
|
for (int i = 0; i < N; i++)
|
|
for (int j = 0; j < M; j++) {
|
|
formula.push_back(OR(NOT(vec[i]), x[std::pair<int,int>(i-1,j-1)], NOT(x[std::pair<int,int>(i,j)])));
|
|
formula.push_back(OR(NOT(vec[i]), NOT(x[std::pair<int,int>(i-1,j-1)]), x[std::pair<int,int>(i,j)]));
|
|
formula.push_back(OR(vec[i], x[std::pair<int,int>(i-1,j)], NOT(x[std::pair<int,int>(i,j)])));
|
|
formula.push_back(OR(vec[i], NOT(x[std::pair<int,int>(i-1,j)]), x[std::pair<int,int>(i,j)]));
|
|
#if 0
|
|
// explicit resolution clauses -- in tests it was better to let the sat solver figure those out
|
|
formula.push_back(OR(NOT(x[std::pair<int,int>(i-1,j-1)]), NOT(x[std::pair<int,int>(i-1,j)]), x[std::pair<int,int>(i,j)]));
|
|
formula.push_back(OR(x[std::pair<int,int>(i-1,j-1)], x[std::pair<int,int>(i-1,j)], NOT(x[std::pair<int,int>(i,j)])));
|
|
#endif
|
|
}
|
|
|
|
for (int j = 0; j < M; j++) {
|
|
if (j+1 <= min_hot)
|
|
formula.push_back(x[std::pair<int,int>(N-1,j)]);
|
|
else if (j+1 > max_hot)
|
|
formula.push_back(NOT(x[std::pair<int,int>(N-1,j)]));
|
|
}
|
|
|
|
return expression(OpAnd, formula);
|
|
}
|
|
#else
|
|
static std::vector<int> lfsr_sym(ezSAT *that, const std::vector<int> &vec, int poly)
|
|
{
|
|
std::vector<int> out;
|
|
|
|
for (int i = 0; i < int(vec.size()); i++)
|
|
if ((poly & (1 << (i+1))) != 0) {
|
|
if (out.empty())
|
|
out.push_back(vec.at(i));
|
|
else
|
|
out.at(0) = that->XOR(out.at(0), vec.at(i));
|
|
}
|
|
|
|
for (int i = 0; i+1 < int(vec.size()); i++)
|
|
out.push_back(vec.at(i));
|
|
|
|
return out;
|
|
}
|
|
|
|
static int lfsr_num(int vec, int poly, int cnt = 1)
|
|
{
|
|
int mask = poly >> 1;
|
|
mask |= mask >> 1;
|
|
mask |= mask >> 2;
|
|
mask |= mask >> 4;
|
|
mask |= mask >> 8;
|
|
mask |= mask >> 16;
|
|
|
|
while (cnt-- > 0) {
|
|
int bits = vec & (poly >> 1);
|
|
bits = ((bits & 0xAAAAAAAA) >> 1) ^ (bits & 0x55555555);
|
|
bits = ((bits & 0x44444444) >> 2) ^ (bits & 0x11111111);
|
|
bits = ((bits & 0x10101010) >> 4) ^ (bits & 0x01010101);
|
|
bits = ((bits & 0x01000100) >> 8) ^ (bits & 0x00010001);
|
|
bits = ((bits & 0x00010000) >> 16) ^ (bits & 0x00000001);
|
|
vec = ((vec << 1) | bits) & mask;
|
|
}
|
|
|
|
return vec;
|
|
}
|
|
|
|
int ezSAT::manyhot(const std::vector<int> &vec, int min_hot, int max_hot)
|
|
{
|
|
// many-hot encoding using LFSR as counter
|
|
|
|
int poly = 0;
|
|
int nbits = 0;
|
|
|
|
if (vec.size() < 3) {
|
|
poly = (1 << 2) | (1 << 1) | 1;
|
|
nbits = 2;
|
|
} else
|
|
if (vec.size() < 7) {
|
|
poly = (1 << 3) | (1 << 2) | 1;
|
|
nbits = 3;
|
|
} else
|
|
if (vec.size() < 15) {
|
|
poly = (1 << 4) | (1 << 3) | 1;
|
|
nbits = 4;
|
|
} else
|
|
if (vec.size() < 31) {
|
|
poly = (1 << 5) | (1 << 3) | 1;
|
|
nbits = 5;
|
|
} else
|
|
if (vec.size() < 63) {
|
|
poly = (1 << 6) | (1 << 5) | 1;
|
|
nbits = 6;
|
|
} else
|
|
if (vec.size() < 127) {
|
|
poly = (1 << 7) | (1 << 6) | 1;
|
|
nbits = 7;
|
|
} else
|
|
// if (vec.size() < 255) {
|
|
// poly = (1 << 8) | (1 << 6) | (1 << 5) | (1 << 4) | 1;
|
|
// nbits = 8;
|
|
// } else
|
|
if (vec.size() < 511) {
|
|
poly = (1 << 9) | (1 << 5) | 1;
|
|
nbits = 9;
|
|
} else {
|
|
assert(0);
|
|
}
|
|
|
|
std::vector<int> min_val;
|
|
std::vector<int> max_val;
|
|
|
|
if (min_hot > 1)
|
|
min_val = vec_const_unsigned(lfsr_num(1, poly, min_hot), nbits);
|
|
|
|
if (max_hot >= 0)
|
|
max_val = vec_const_unsigned(lfsr_num(1, poly, max_hot+1), nbits);
|
|
|
|
std::vector<int> state = vec_const_unsigned(1, nbits);
|
|
|
|
std::vector<int> match_min;
|
|
std::vector<int> match_max;
|
|
|
|
if (min_hot == 1)
|
|
match_min = vec;
|
|
|
|
for (int i = 0; i < int(vec.size()); i++)
|
|
{
|
|
state = vec_ite(vec[i], lfsr_sym(this, state, poly), state);
|
|
|
|
if (!min_val.empty() && i+1 >= min_hot)
|
|
match_min.push_back(vec_eq(min_val, state));
|
|
|
|
if (!max_val.empty() && i >= max_hot)
|
|
match_max.push_back(vec_eq(max_val, state));
|
|
}
|
|
|
|
int min_matched = min_hot ? vec_reduce_or(match_min) : CONST_TRUE;
|
|
int max_matched = vec_reduce_or(match_max);
|
|
|
|
return AND(min_matched, NOT(max_matched));
|
|
}
|
|
#endif
|
|
|
|
int ezSAT::ordered(const std::vector<int> &vec1, const std::vector<int> &vec2, bool allow_equal)
|
|
{
|
|
std::vector<int> formula;
|
|
int last_x = CONST_FALSE;
|
|
|
|
assert(vec1.size() == vec2.size());
|
|
for (size_t i = 0; i < vec1.size(); i++)
|
|
{
|
|
int a = vec1[i], b = vec2[i];
|
|
formula.push_back(OR(NOT(a), b, last_x));
|
|
|
|
int next_x = i+1 < vec1.size() ? literal() : allow_equal ? CONST_FALSE : CONST_TRUE;
|
|
formula.push_back(OR(a, b, last_x, NOT(next_x)));
|
|
formula.push_back(OR(NOT(a), NOT(b), last_x, NOT(next_x)));
|
|
last_x = next_x;
|
|
}
|
|
|
|
return expression(OpAnd, formula);
|
|
}
|
|
|